echo1-coco-split provides a faster, safer way to split coco formatted datasets into train, validation and test sets.
# Install echo1-coco-split
pip install echo1-coco-split
# Run the coco-split
coco-split \
--has_annotations \
--valid_ratio .15 \
--test_ratio .05 \
--annotations_file ./instances_default.json
Saved 490 annotations from 53 images to train.json
Saved 136 annotations from 11 images to valid.json
Saved 34 annotations from 4 images to test.json
usage: coco-split [-h] --annotations_file ANNOTATIONS_FILE --valid_ratio VALID_RATIO --test_ratio
TEST_RATIO [--train_name TRAIN_NAME] [--valid_name VALID_NAME]
[--test_name TEST_NAME] [--has_annotations] [--seed SEED]
Splits a coco annotations file into a training, validation, and test set.
options:
-h, --help show this help message and exit
--annotations_file ANNOTATIONS_FILE
Path to COCO annotations file.
--valid_ratio VALID_RATIO
Set valid dataset ratio
--test_ratio TEST_RATIO
Set test dataset ratio
--train_name TRAIN_NAME
Where to store COCO training annotations
--valid_name VALID_NAME
Where to store COCO valid annotations
--test_name TEST_NAME
Where to store COCO test annotations
--has_annotations Ignore all images without annotations. Keep only those with at least one
annotation
--seed SEED Set the seed generator
Previous work done by: